Which weighting scheme for inverse probability of treatment weighting (IPTW) in the presence of extreme weights or propensity score imbalance?
Objectifs
Inverse probability of treatment weighting (IPTW) is a reference method for balancing covariates between treatment groups and estimating unbiased causal effects in observational studies. However, high or extreme weights can lead to imprecise effect estimates and excessive variance, particularly problematic at the extremes of the propensity score (PS) distribution. This work aims to compare different IPTW weight management strategies according to the targeted estimand (ATE, ATT, ATO) and their impact on covariate balance, effect estimates, and target population interpretation.
Conclusion
IPTW creates balanced pseudo-populations and serves as foundation for advanced methods incorporating time-varying weights. Weight calculation methods must be transparently described and adapted to study context. Appropriate weighting scheme selection optimizes treatment effect estimation and yields more robust causal estimates, with ATO estimands showing greater stability than ATE in the presence of extreme PS values.